ChatGPT: Curse or Blessing for Democracy?

In this episode of our Swiss AI podcast we tackle a charged topic: the influence of ChatGPT on democracy. This episode dives deep into the complex interplay between artificial intelligence and democratic discourse. We discuss both the potential dangers and the possible benefits that ChatGPT and similar technologies can bring to democratic processes. Along the way we examine topics such as opinion manipulation, information dissemination, and the role of AI in political decision-making. Our goal is to develop a balanced understanding of how these innovative technologies can both challenge and enrich our democracy.

Cambridge Analytica: The Big Bang of Digital Campaigning

Let us start with a look back at the Cambridge Analytica scandal of 2016. For the younger among you: that was when Facebook was still cool and we all thought our data was safe. Spoiler: it was not.

Cambridge Analytica used Facebook data to build detailed personality profiles and deploy them for political microtargeting. Sounds complicated? It is. Essentially they figured out that you like cat videos and therefore showed you Trump ads. To put it simply.

"Who would have thought that your fondness for cat memes would turn you into a voter?" - Nobody at Facebook, apparently.

The methods of the time look like stone-age technology compared to what is possible today. Modern language models like GPT-4 enable a precision and scalability in personalised communication that would give even George Orwell cold sweats.

AI: Master of Deception

A central theme in our discussion was the growing difficulty of telling real from fake. AI systems can now imitate text, images, and even voices so convincingly that even my mother, who normally sees through every trick, would be taken in.

This becomes particularly acute in a political context. When voters can no longer tell whether a statement actually came from a politician or was generated by AI, we have a problem. A big problem.

"In a world where AI can imitate Donald Trump, who still needs the real Donald Trump?" - Probably nobody, but that is a different topic.

OpenAI's Well-Intentioned but Ineffective Attempt

OpenAI's announcement to block its chatbot for certain forms of electoral manipulation is roughly as effective as trying to hold back a flood with a tea strainer. Well-intentioned, but ultimately pointless.

As we discussed on the podcast, such restrictions are easy to circumvent. Whether through clever prompts or the use of alternative, less restrictive models. It is like trying to make the internet child-safe by putting up a sign: "Please do not view inappropriate content." Good luck with that.

The Unexpected Democratisation through AI

Despite all the risks, and there are many, the AI revolution also offers opportunities for democracy. The same technologies that can be used for manipulation also enable a degree of transparency and participation that was previously unimaginable.

AI-powered analysis tools could allow citizens to understand complex political topics more readily. Imagine being able to actually understand what a piece of legislation says without a law degree. Revolutionary, I know.

At the same time politicians could use AI to develop a deeper understanding of the needs and concerns of their voters. Though, looking at some politicians, perhaps any form of understanding would already be progress.

The Regulatory Nightmare

A large part of our discussion revolved around the question of how AI in political contexts could be regulated. Spoiler: it is complicated. Very complicated.

The EU AI Act is an ambitious attempt to create a regulatory framework for AI systems. But as we discussed on the podcast, there are already considerable difficulties at the level of defining what is being regulated. What exactly is "AI"? Where do we draw the line between conventional software and AI systems?

"Define AI," said the legislator. "Define intelligence," replied the AI. And so a philosophical discourse began that continues to this day.

These questions are not just academic sparring. They have far-reaching consequences for the practical feasibility and effectiveness of regulations. A definition of AI that is too narrow could let problematic applications slip through the net. One that is too broad could unnecessarily hamper innovation. It is like trying to regulate water with a sieve.

Education: The Underrated Hero

An often overlooked but decisive factor in this equation is digital literacy. As we worked out on the podcast, general understanding of the possibilities and risks of AI systems lags behind technological development. It is as if we were teaching stone-age people how to use a smartphone.

The challenge is to make broad sections of the population, especially older generations, aware of the nuances of the digital information landscape. And no, "Have you tried turning it off and on again?" is not adequate digital education.

Ethics: The Philosophical Minefield

Our discussion also touched on profound ethical questions. When AI systems are able to generate individualised political messages tailored precisely to the fears, hopes, and prejudices of individual voters, where do we draw the line between legitimate political communication and manipulation?

"If an AI tells you what you want to hear, is it still democracy or already technocracy?" - A question that would even make Socrates ponder.

It is tempting to view these technologies as neutral and to place responsibility solely with the users. But that would be too simple. The architecture and training of AI systems contain inherent value judgements and biases. It is like teaching a child what is right and wrong, except this child potentially has the power to influence elections.

The Future: A Glance into the Crystal Ball (or Rather the Quantum Computer)

Finally, we ventured a look into the future. The development of Artificial General Intelligence (AGI) could take the discussion about the influence of AI on democratic processes into an entirely new dimension. If AI systems surpass human intelligence in all areas, what role do human decision-makers still play in political processes?

Imagine we had an AI as Federal Chancellor. No sleep, no scandals, no embarrassing dance routines at folk festivals. Sounds tempting, right? But who programmes the values of this AI? And who programmes the programmers?

It Remains Complicated

Our discussion on the Attention Heads podcast has shown that the challenges AI poses to democratic processes are multi-layered and complex. There are no simple solutions, but there are some clear imperatives for action:

We need to increase the transparency and traceability of AI systems in political contexts. And no, "It is magic" is not an acceptable explanation. Regulatory frameworks must be designed to be flexible and forward-looking. So please no laws that are already out of date before the ink is dry. Digital literacy must become a societal priority. And that means going beyond "How do I post a selfie on Instagram?" Ethical considerations must be integrated from the outset into the development and deployment of AI systems. Yes, even if that means giving philosophers a job. The future of our democracy in the digital age will depend on how well we as a society master these challenges. It is up to all of us, technologists, politicians, education experts, and citizens, to develop solutions together that protect the integrity of our democratic processes while harnessing the positive potential of AI.

The discussion is open, and it is more important than ever. Let us conduct it with the seriousness it deserves. But please without the panic. Because if there is one thing AI still cannot do, it is replace our sense of humour. At least I hope so.